NMFk Source Identification via Semi-Supervised Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current technologies face challenges in identifying and separating unknown sources from mixed signals, particularly when signal variations occur due to wave-like propagation, diffusion, or advection, and when the number of sources is unknown, requiring robust and efficient methods to determine source locations and signals in complex environments.
Innovation Solution
The use of Non-negative Matrix Factorization (NMF) combined with semi-supervised clustering, specifically the NMFk and Shift-NMFk procedures, which perform multiple trials to estimate the number of sources and account for signal shifts, using selection criteria to ensure robustness and accuracy in source identification, and incorporating Green's functions for advection-diffusion models in the Green-NMFk method.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional source separation methods are used, then the system is simpler to implement, but the accuracy of source identification deteriorates when signal variations occur due to wave-like propagation, diffusion, or advection
Solution Approach 1:
The patent transforms the source separation problem by changing the parameter representation from direct signal values to factorized non-negative matrices. By representing mixed signals as products of source matrices and mixing matrices under non-negativity constraints, the method adapts to signal variations caused by wave-like propagation, diffusion, and advection while maintaining computational tractability through iterative optimization algorithms.
Solution Approach 2:
The patent introduces non-negative matrix factorization as an intermediary computational framework between the observed mixed signals and the underlying source signals. This intermediary approach uses clustering algorithms to identify source patterns and iterative optimization to separate mixed signals, serving as a bridge that handles complex signal variations without requiring direct complex mathematical transformations.
2Adaptability or versatility
If the number of sources is unknown, then the method can handle more realistic scenarios, but the computational complexity increases due to multiple trials and clustering procedures
Solution Approach 1:
The patent performs preliminary clustering of sensor signals before source separation to identify potential source patterns and estimate the number of sources. By pre-processing the data through clustering algorithms, the method prepares organized source representations that guide subsequent NMF iterations, reducing the need for extensive trial-and-error computations and accelerating convergence when the number of sources is unknown.
Solution Approach 2:
The patent implements feedback mechanisms where clustering results from previous iterations inform the NMF decomposition parameters for subsequent iterations. The algorithm continuously evaluates clustering quality metrics and adjusts the number of sources and factorization parameters based on feedback from data reconstruction errors, enabling adaptive convergence without exhaustive computational trials.
3Reliability
If multiple NMF trials are performed to ensure robustness, then the reliability of source identification improves, but the processing time increases
Solution Approach 1:
The patent performs a limited number of NMF trials with different initializations rather than exhaustive searches. By conducting partial trials (e.g., 5-10 random initializations) and using clustering to guide selection of promising solutions, the method achieves sufficient robustness without the computational burden of exhaustive trials, balancing reliability with processing efficiency.
Solution Approach 2:
The patent replaces brute-force multiple trial computations with intelligent clustering-guided selection. Instead of mechanically executing numerous NMF trials and evaluating all results, the system uses clustering algorithms to identify representative source patterns and selects only the most promising trials for full NMF decomposition, substituting computational brute force with algorithmic intelligence.
4Measurement precision
If signal propagation models are incorporated, then the accuracy in complex environments improves, but the device complexity and computational requirements increase
Solution Approach 1:
The patent develops a universal NMF framework that can accommodate multiple signal propagation models (wave-like propagation, diffusion, advection) through a unified mathematical structure. By formulating the problem in terms of non-negative matrix factorization with general mixing matrices, the system can adapt to different propagation physics without requiring separate specialized algorithms for each model type, reducing overall system complexity while maintaining accuracy.
Solution Approach 2:
The patent handles different signal propagation models by changing the parameterization of the mixing matrix rather than changing the fundamental algorithm. For wave-like propagation, the mixing matrix incorporates delay parameters; for diffusion, it incorporates spreading parameters; for advection, it incorporates transport parameters. This parameter-based adaptation allows a single NMF framework to handle diverse propagation physics without increasing structural complexity.
Data Source
AI summary
Machine-learning methods and apparatus are provided to solve blind source separation problems with an unknown number of sources and having a signal propagation model with features such as wave-like propagation, medium-dependent velocity, attenuation, diffusion, and/or advection, between sources and sensors. In exemplary embodiments, multiple trials of non-negative matrix factorization are performed for a fixed number of sources, with selection criteria applied to determine successful trials. A semi-supervised clustering procedure is applied to trial results, and the clustering results are evaluated for robustness using measures for reconstruction quality and cluster separation. The number of sources is determined by comparing these measures for different trial numbers of sources. Source locations and parameters of the signal propagation model can also be determined. Disclosed methods are applicable to a wide range of spatial problems including chemical dispersal, pressure transients, and electromagnetic signals, and also to non-spatial problems such as cancer mutation.